Hydrocarbon-like and oxygenated organic aerosols in Pittsburgh: insights into sources and processes of organic aerosols

Transcription

1 Atmos. Chem. Phys., 5, , 2005 SRef-ID: /acp/ European Geosciences Union Atmospheric Chemistry and Physics Hydrocarbon-like and oxygenated organic aerosols in Pittsburgh: insights into sources and processes of organic aerosols Q. Zhang 1,*, D. R. Worsnop 2, M. R. Canagaratna 2, and J. L. Jimenez 1,3 1 Cooperative Institute for Research in Environmental Sciences (CIRES), 216 UCB, University of Colorado-Boulder, Boulder, Colorado , USA 2 Aerodyne Research Inc., Billerica, Massachusetts , USA 3 Department of Chemistry and Biochemistry, 216 UCB, University of Colorado-Boulder, Boulder, Colorado , USA * now at: Atmospheric Science Research Center, State University of New York, 251 Fuller Rd, Albany, NY 1220, USA Received: 8 August 2005 Published in Atmos. Chem. Phys. Discuss.: 9 September 2005 Revised: 7 December 2005 Accepted: 7 December 2005 Published: 1 December 2005 Abstract. A recently developed algorithm (Zhang et al., 2005) has been applied to deconvolve the mass spectra of organic aerosols acquired with the Aerosol Mass Spectrometer (AMS) in Pittsburgh during September The results are used here to characterize the mass concentrations, size distributions, and mass spectra of hydrocarbon-like and oxygenated organic aerosol (HOA and OOA, respectively). HOA accounts for 3% of the measured organic aerosol mass and OOA accounts for 66%. The mass concentrations of HOA demonstrate a prominent diurnal profile that peaks in the morning during the rush hour and decreases with the rise of the boundary layer. The diurnal profile of OOA is relatively flat and resembles those of SO 2 and NH +. The size distribution of HOA shows a distinct ultrafine mode that is commonly associated with fresh emissions while OOA is generally concentrated in the accumulation mode and appears to be mostly internally mixed with the inorganic ions, such as SO 2 and NH +. These observations suggest that HOA is likely primary aerosol from local, combustion-related emissions and that OOA is secondary organic aerosol (SOA) influenced by regional contributions. There is strong evidence of the direct correspondence of OOA to SOA during an intense new particle formation and growth event, when condensational growth of OOA was observed. The fact that the OOA mass spectrum from this event is very similar to that from the entire study suggests that the majority of OOA in Pittsburgh is likely SOA. O 3 appears to be a poor indicator for OOA concentration while SO 2 is a relatively good surrogate for this dataset. Since the diurnal averages of HOA track those of CO during day time, oxidation/aging of HOA appears to be very small on the time scale Correspondence to: J. L. Jimenez (jose.jimenez@colorado.edu) of several hours. Based on extracted mass spectra and the likely elemental compositions of major m/z s, the organic mass to organic carbon ratios (OM:OC) of HOA and OOA are estimated at 1.2 and 2.2 µg/µgc, respectively, leading to an average OM:OC ratio of 1.8 for submicron OA in Pittsburgh during September. The C:O ratio of OOA is estimated at 1:0.8. The carbon contents in HOA and OOA estimated accordingly correlate well to primary and secondary organic carbon, respectively, estimated by the OC/EC tracer technique (assuming POC-to-EC ratio=1). In addition, the total carbon concentrations estimated from the AMS data agree well with those measured by the Sunset Laboratory Carbon analyzer (r 2 =0.87; slope=1.01±0.11). Our results represent the first direct estimate of the OM:OC ratio from highly timeresolved chemical composition measurements. 1 Introduction Organic compounds are ubiquitous and abundant in ambient aerosols. They typically account for 20 50% of the fine particle mass (Jacobson et al., 2000; Kanakidou et al., 2005; NARSTO, 2003; Saxena and Hildemann, 1996; Seinfeld and Pankow, 2003) and are often internally mixed in the same particles with inorganic aerosols (Middlebrook et al., 2003, 1998; Murphy et al., 1998). Organic compounds play important roles in the formation, growth, and removal of ambient aerosols (IPCC, 2001). They also significantly affect the hygroscopicity (Saxena et al., 1995), toxicity (Sheesley et al., 2005), direct radiative properties (Chung and Seinfeld, 2002; Haywood and Boucher, 2000), and indirect effects (Facchini et al., 1999) of atmospheric aerosols and therefore have major implications for climate, visibility, and human health Author(s). This work is licensed under a Creative Commons License.

2 3290 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Elucidating the urban-to-global roles as well as the sources and fate of atmospheric aerosols inherently must rely on a thorough understanding of the chemical and microphysical properties of particulate organics. However, it is extremely difficult to obtain a complete description of the molecular composition of aerosol organics because of the number, complexity, and extreme range of physical and chemical properties of these compounds. Usually analysis of over a hundred different molecules can only account for 10 20% of the organic mass (NARSTO, 2003; Rogge et al., 1993). For these reasons in order to understand the chemistry of atmospheric organic aerosols, bulk characterization approaches such as those targeting compound classes and/or bulk properties (Fuzzi et al., 2001; Gelencser, 200; Murphy, 2005) should be developed in addition to compound-specific techniques. Spectroscopic techniques, including Fourier transform infrared (FTIR) spectroscopy (Allen et al., 199; Blando et al., 1998, 2001; Edney et al., 2003; Laurent and Allen, 200; Maria et al., 2002; Russell, 2003) and nuclear magnetic resonance (NMR) (Decesari et al., 2005, 2000; Fuzzi et al., 2001), have been applied to characterize the functional group composition of aerosol organics. A major advantage of these techniques is that they characterize ambient aerosols based on the majority of the organic mass, rather than a limited number of molecules (Allen et al., 199; Blando et al., 1998; Decesari et al., 2000; Fuzzi et al., 2001; Maria et al., 2002). However, in analysis of ambient samples, both FTIR and NMR methods rely on assumptions about the relationship between the strength of the electromagnetic interaction and the amount of material that may introduce significant uncertainties in quantification of functional groups (Blando et al., 2001; Fuzzi et al., 2001). In addition, until now neither method has been adapted for real-time sampling, nor are they capable to determine ambient organic aerosols with high time and size resolution. Mass spectrometry techniques have been widely used in aerosol analysis because of their universal, extremely sensitive, and rapid detection of aerosol components (Jayne et al., 2000; Jimenez, 2005; Johnston, 2000; McKeown et al., 1991; Murphy, 2005; Noble and Prather, 2000; Suess and Prather, 1999). Among these, the Aerodyne Aerosol Mass Spectrometer (AMS) (Jayne et al., 2000; Jimenez et al., 2003) is the most commonly used. It is capable of quantitatively measuring the size-resolved mass concentrations of organic aerosols with a time resolution of minutes (e.g., Allan et al., 2003a; Drewnick et al., 200a, b; Jimenez et al., 2003; Zhang et al., 2005b). Good correlations between the mass concentrations of organic aerosols measured by an AMS and the organic carbon concentrations measured by thermal-optical Carbon Analyzers have been observed in various locations, including Pittsburgh (Zhang et al., 2005b), Houston (Canagaratna et al., ), Tokyo (Takegawa et al., 2005b), and the coast 1 Canagaratna, M., Jimenez, J. L., Silva, P., et al.: Time resolved of New England (Bates et al., 2005; de Gouw et al., 2005). The AMS employs thermal vaporization (usually at 600 C) and 70 ev electron ionization that generally causes extensive fragmentation of organic molecules (Alfarra, 200; Jayne et al., 2000; Jimenez et al., 2003). As a result, in ambient analysis each mass-to-charge ratio (m/z) peak in an AMS mass spectrum may contain contributions from many different molecules. For this reason the AMS does not characterize individual molecules in ambient air, but rather the methodology fits into the group of techniques that characterizes bulk chemistry of organic aerosols. A recently developed custom principal component analysis technique makes it possible to use an AMS to identify and quantify broad aerosol classes that have different temporal and mass spectral signatures (Zhang et al., 2005a). When applied in urban areas, this technique deconvolves and quantifies two types of organic aerosols, hydrocarbon-like and oxygenated (HOA and OOA, respectively), which together account for almost all the organic aerosol mass measured by the AMS (Zhang et al., 2005a). Hydrocarbon-like aerosols are named based on the similarity of their AMS mass spectra to those of hydrocarbons mixtures, while oxygenated organic aerosols are named based on their high oxygen content (Zhang et al., 2005a). More importantly, this technique allows the extraction of mass concentrations, size distributions, and mass spectra of HOA and OOA that are physically and chemically meaningful. As reported by Zhang et al. (2005a), the extracted mass spectrum of HOA is remarkably similar to the spectra of directly sampled vehicle exhaust and lab-generated lubricating oil aerosols, while the spectrum of OOA closely resembles those of highly processed organic aerosols sampled at rural and remote locations. The OOA spectrum also shows similarity with that of fulvic acid (Alfarra, 200; Zhang et al., 2005a) a humiclike substance that is ubiquitous in the environment and has previously been used as an analogue to represent polyacid components found in highly processed and oxidized atmospheric organic aerosols (Decesari et al., 2002). In this paper we report the application of this technique to the AMS data acquired at the U.S. EPA Pittsburgh Supersite and the major findings regarding the time trends, concentrations, and size distributions of HOA and OOA in Pittsburgh. These results complement two earlier publications from us that discuss the chemistry of new particle growth (Zhang et al., 200) and the general characteristics of submicron aerosol species (inorganic ions + organics) in Pittsburgh (Zhang et al., 2005b). Because those two articles were published before the development of the HOA and OOA deconvolution technique (Zhang et al., 2005a) discussions concerning aerosol organics in both were made based on analysis of total organic signals and a few m/z fragments. Only in this current paper are we able to provide an in-depth analysis of aerosol size and chemical composition measured during the Texas Air Quality Study, in preparation, Atmos. Chem. Phys., 5, , 2005

3 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols 3291 the possible sources and processes of organic aerosols based on 1) correlations of HOA and OOA with gas phase and organic carbon measurements; 2) temporal variations, size distributions, and mass spectra of HOA and OOA; and 3) the dynamics of HOA and OOA during an intense new particle formation and growth event. 2 Experimental and data analysis methods The AMS data used for this study were acquired during 7 22 September 2002 from the main site of the Pittsburgh Air Quality Study (PAQS). Dates and times are reported in Eastern Standard Time (EST). The local time during this study was Eastern Daylight Savings Time (EDT), which is 1 h ahead of EST. An overview on the sampling location, instrumentation, and the objectives of PAQS is given elsewhere (Wittig et al., 200). Gas-phase and meteorological variables were measured simultaneously (Wittig et al., 200). Note that the original CO data of this study period were offset by 0.35 ppm to adjust the average minimum CO concentration during periods of very clean air (e.g., air masses from the north) to 0.1 ppm, which is the background concentration of CO in Northern Hemisphere (Finlayson-Pitts and Pitts, 2000). 2-h averaged PM 2.5 EC and OC were measured in situ using a Sunset Laboratory thermal optical transmittance carbon analyzer (sampling details are given by Polidori et al., ). Detailed information on the AMS operation and data analysis is presented by Zhang et al. (2005b, 200). The mass concentrations and size distributions of fine particle species (e.g., SO 2, NO 3, NH+, and organics) measured by the AMS during this study compare well with measurements made by collocated instruments, with some systematic differences due to different size cuts (Zhang et al., 2005b). The absolute accuracy of the data reported here is mainly limited by the uncertainties in AMS particle collection efficiency (Zhang et al., 2005b). The mass concentrations and mass spectra of HOA and OOA were derived using the deconvolution procedures described in a separate publication (Zhang et al., 2005a). This technique involves a series of multivariate linear regressions that use mass-to-charge ratios (m/z s) 57 (mostly C H + 9 ) and (mostly CO + 2 ), the identified AMS mass spectral tracers for HOA and OOA, respectively, as the time series of the initial principal components followed by an iterative algorithm to determine HOA and OOA time series and mass spectra. The time resolution of the HOA and OOA time series are 5 10 min. Because of the use of a quadrupole mass spectrometer only a subset of m/z s (16 in total, out of which 8 are mainly organic m/z s) were scanned for size distributions in this study (Jayne et al., 2000; Jimenez et al., 2003; Zhang et al., 2005b). We are thus unable to derive the size distributions of HOA and OOA using the full mass spectra. Those presented in this study are derived based on the measured size distributions of m/z s 57 and because they are the first order AMS tracers for OOA and HOA, respectively, and correlate closely to the HOA and OOA time series (Zhang et al., 2005a). The size distribution of OOA was derived by normalizing the integrated signals of m/z in nm particles to the estimated concentrations of OOA. Note that this size range was selected to ensure the capture of all particle signals. The first paper on the AMS showed about 50% transmission at 1000 nm, which is typically referred to as PM 1 (Fig. 9 on Jayne et al., 2000) and a detailed analysis of signals from Pittsburgh AMS data showed partial transmission for particles down to 33 nm (Zhang et al., 2005a). Slightly different aerodynamic lens designs have been used in the AMS, including during this study, and each lens may have slightly different particle transmission characteristics. This is a subject of ongoing research that will be presented in a future publication. The potential for interferences to m/z to cause differences between the real size distribution of OOA and the distribution presented here is low because of the very high correlation of m/z and OOA and the lack of m/z signal in the HOA mass spectrum (Zhang et al., 2005a). However, considering that m/z 57 is present in the OOA mass spectrum at intensity 2% of that of m/z, which indicates that m/z 57 may have contributions from oxygenated species (e.g., C 3 H 5 O + ) in addition to hydrocarbons (i.e., C H + 9 ), we derived the size distribution of HOA by subtracting 2% of the m/z signal from the size distribution of m/z 57 and then normalizing the integrated signals in nm particles to the estimated concentrations of HOA. The presence of oxygenated m/z 57 (C 3 H 5 O + ) when the OOA to HOA ratio is high has been confirmed by recently acquired high m/z resolution AMS data. Organic mass to organic carbon ratios (OM:OC) of HOA and OOA are estimated based on the extracted mass spectra of the two components (Zhang et al., 2005a) and the likely elemental compositions of the major m/z s in the corresponding spectrum. All the data reduction and analysis are performed with Igor Pro 5 (Wavemetrics Inc.). 2 Polidori, A., Turpin, B. J., Lim, H.-J., Cabada, J. C., Subramanian, R., Robinson, A. L., and Pandis, S. N.: Local and regional secondary organic aerosol: Insights from a year of semi-continuous carbon measurements at Pittsburgh, Aerosol Sci. Technol., submitted, Atmos. Chem. Phys., 5, , 2005

4 Figure Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Mass Concentration (µg m -3 ) 6 e HOA OOA % of Total Organics 100 f HOA OOA 90% 75% Median Mean 25% 10% Fig. 1. (a) Time series of the absolute and fractional HOA and OOA in Pittsburgh during 7 22 September Missing data points are due to either occasional instrumental malfunction or maintenance/calibration. Average diurnal cycles of the mass concentrations of (b) HOA and (c) OOA. (d) Average diurnal cycles of the fractional contribution of HOA and OOA to the total organic aerosol mass. Box plots of (e) the mass concentration and (f) the fractional contribution of HOA and OOA to total organics. The box plots are read as follows: the upper and lower boundaries of the box indicate the 75th and the 25th percentiles, the solid line within the box marks the median, the whiskers above and below the box indicate the 90th and 10th percentiles. Cross symbols or the red broken lines represent the means. The results of the statistical analysis are given in Table 1. The x-axis labels of the diurnal plots corresponds to the hour that ends the averaging interval and the ordinal of the hour in the day, e.g., 1 means the first hour of the day, from 00:00 01:00 a.m. EST. 3 Results and discussion 3.1 Mass concentrations and temporal variations of HOA and OOA Mass concentrations and diurnal variations of HOA and OOA The time series of the mass concentration and the fractional contribution of hydrocarbon-like and oxygenated organic aerosols are shown in Fig. 1 (and see Zhang et al., Atmos. Chem. Phys., 5, , a). During this study in Pittsburgh the mass concentrations of atmospheric fine particles changed dramatically. Multiday episodes of fine particle pollution are interleaved with clean periods following heavy rainfall and/or the arrival of clean air from the north. The time trends of HOA and OOA are very different, except for a few periods when their concentrations appear to co-vary due to the arrival of clean air masses and/or rainfall scavenging. In general, the time series of HOA demonstrates a pronounced variation pattern that typically peaks during morning rush hours, when the mixing layer is relatively shallow and primary emissions from traffic 1

5 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols 3293 Figure 2. Fig. 2. Time trends of (a) HOA and typical combustion emission tracers (CO, NO x, EC), (b) OOA and PM 1 SO 2 (both from the AMS), and (c) the OOA to organic mass ratio and O 3 during 7 22 September 2002 in Pittsburgh. are intense. OOA demonstrates a time trend similar to that of sulfate (Fig. 2b), a dominant secondary aerosol species that is strongly influenced by regional accumulation rather than local emissions in Pittsburgh (Zhang et al., 2005b). The significantly different diurnal patterns of HOA and OOA are evident in Figs. 1b and c. Note that these diurnal averages may be skewed by a few abnormally low/high loading events due to the relatively short duration of this study (16 days). The dip at the 18th hour (between 5:00 to 6:00 p.m.) on the diurnal curve of OOA, for example, is mainly caused by the abrupt drop in the mass concentration associated with a rainfall event in the afternoon of 15 September. HOA demonstrates a clear diurnal pattern that peaks in the morning during the rush hour (8:00 9:00 a.m.), gradually decreases after 8:00 a.m., and reaches its minimum between 3:00 5:00 p.m. In contrast, the OOA diurnal profile is relatively flat and resembles those of SO 2 and NH + (Zhang et al., 2005b). In addition, while the trend is relatively weak, the mean values of OOA show slight increases in the afternoon between 13:00 to 16:00 EST (Fig. 1c), when photochemistry is relatively intense. As a result, the highest fraction of OOA was observed in the afternoon around 15:00 17:00 EST, during which OOA accounts for more than 80% of the total organic mass on average (Fig. 1d). Table 1 and Figs. 1e, f summarize the statistics of the mass concentrations of HOA and OOA and their fractional contributions to the total organics. OOA dominates organic aerosol mass loading in Pittsburgh, accounting for more than half of the organic mass for 85% of the time during this Table 1. Statistical parameters of the distributions of the mass concentrations of HOA and OOA, and of their fractional contributions to the total organic mass in Pittsburgh during 7 22 September Mass Concentration (µg m 3 ) % of Total Organics a HOA OOA HOA OOA Mean σ Median Min <D.L. b <D.L. b 0 7 Max th percentile th percentile th percentile th percentile a These values are obtained by analyzing the relative concentrations (percent values) of HOA and OOA, rather than estimated from the absolute mass concentration statistics. b D.L.: Detection limit of organic mass concentration, which was estimated to be 0.15 µg m 3 for this study (Zhang et al., 2005b). study (Fig. 1a). The average (±1σ ) mass concentration of OOA is 2.93 (±1.65) µg m 3, roughly twice that of HOA (1.8±1. µg m 3 ; Figs. 1e, f). On average, HOA represents 3% of the organic aerosol mass in Pittsburgh while OOA accounts for 66%. Even during the morning rush hour, the mass loading of OOA is larger than that of HOA on average (Fig. 1d). 2 Atmos. Chem. Phys., 5, , 2005

6 329 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Figure 3 Fig. 3. Scatter plots of the concentrations of (a) HOA vs. CO; (b) HOA vs. NO x ; (c) HOA vs. elemental carbon; (d) OOA vs. sulfate; and (e) OOA vs. O 3 (data points are colored by hour of day) Correlation of HOA and OOA with combustion and secondary aerosol tracers Figure 2a shows the time series of HOA together with three primary combustion emission tracers CO, NO x, and elemental carbon (EC). Figure 2b shows the time series of OOA and sulfate a secondary aerosol species that is mainly formed through gas-phase and aqueous-phase oxidation of SO 2. The corresponding linear regression scatter plots are shown in Figs. 3a d. HOA correlates well with CO (r 2 =0.73), NO x (r 2 =0.82), and EC (r 2 =0.72), all of which demonstrate a pronounced diurnal pattern that peaks in the morning when traffic emissions are intense, declines with the rise of the mixed layer depth, and gradually increases after the boundary layer collapses in the evening. Such diurnal behavior is characteristics for air pollutants from local emissions and thus indicates a strong association of HOA to combustion aerosol emitted locally (e.g., from traffic). This hypothesis is consistent with the size distribution of HOA, which constantly shows a prominent ultrafine mode that is common for combustion aerosols (see Sect. 3.2). The mass spectrum of HOA is also very similar to those of freshly emitted vehicle exhaust aerosols, showing ion series characteristic of hydrocarbons (see Sect. 3.3 and Zhang et al., 2005a). The time trend of OOA tracks that of SO 2 (r2 =0.7) but correlates very weakly to the combustion tracers (r 2 <0.1). The good correlation between OOA and SO 2 suggests similar sources and/or processes of these two aerosol components. SO 2 is a major fine particulate species in Pittsburgh due to the high SO 2 emissions in this geographical region (Wittig et al., 200; Zhang et al., 2005b). The atmospheric concentration of SO 2 is strongly influenced by regional accumulation rather than local production since a significant fraction of the fine particles in Pittsburgh are aged over regional scales (Anderson et al., 200; Tang et al., 200). For these reasons the rather weak response of the ambient concentrations of OOA and SO 2 to the daily fluctuation of mixed layer depth is indicative of the regional nature of both components (i.e., similar levels of SO 2 and OOA in the morning boundary layer and in air aloft). This is in contrast to HOA, which originates predominantly from local emissions and as expected demonstrates a pronounced diurnal pattern that peaks in the morning when traffic emissions are high and the mixed layer depth is low. The lack of strong diurnal variations of SO 2 and OOA might also be the result of their relatively high background concentrations in the region daily photochemical production of these two components, which is usually most intense in the afternoon, tends to be dwarfed by the much stronger Atmos. Chem. Phys., 5, ,

7 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols 3295 variations in mass concentrations associated with changes of air mass or rainfall scavenging. Note that in areas where fine particles are more strongly influenced by local photochemistry, such as in Mexico City, photochemical production of oxygenated organics is sufficiently pronounced that a clear increasing trend of OOA is often observed during morning and early afternoon 3. OOA in Pittsburgh appears to be primarily secondary organic aerosol associated with regional accumulation rather than from local emissions. This hypothesis is consistent with the size distributions of OOA, which are dominated by the accumulation mode (see Sect. 3.2), and its mass spectrum, which closely resembles those of aged and highly oxidized organic aerosols (see Sect. 3.2 and Zhang et al., 2005a) Correlation of OOA to O 3 A previous study in Pittsburgh reported the use of ozone as an indicator for SOA formation supported by the observation that increases in the OC-to-EC ratio correlate with ozone increases (Cabada et al., 200a). However, we found very little correlation between O 3 and OOA during this study (r 2 0; Fig. 3e), which suggests that ozone concentration is a rather poor indicator for SOA concentration, at least for this study period. A possible explanation for the positive correlation of SOA to O 3 observed by Cabada et al. (200a) is that the POC/EC ratios used in their EC/OC tracer method were overestimated, possibly by an average factor of 2 as they were determined based on measured ambient OC/EC ratios during periods when ambient organic aerosol likely contained 50% SOA (see discussions in Sect. 3.). Such underestimation of SOA (and overestimation of POA) would greatly diminish as the boundary layer rises, due to the strong dilution of EC. Since the O 3 diurnal profile is anti-correlated to that of EC (due to the strong effect of the boundary layer on both), this would result in an apparent correlation between SOA and O 3. A related effect is illustrated in Fig. 2c where the organicsto-hoa ratio shows a pronounced daytime increase pattern that is similar to ozone (r 2 =0.38). Since the AMS cannot measure EC, the organics-to-hoa ratio is presented as a surrogate for the OC-to-EC ratio given the good correlation between HOA and EC (Fig. 2a) and between organic mass and OC concentrations (r 2 =0.88) (Zhang et al., 2005b). The observed daytime increase of organics-to-hoa ratio (as well as OC-to-EC ratio) is mainly driven by the strong diurnal variations in HOA (and EC) concentrations associated with daily fluctuation of the boundary layer height (Fig. 1b), rather than production of OOA. Due to the good correlation between SO 2 and OOA (r 2 =0.7; Figs. 2b, 3d) as well as the fact that both are in 3 Dzepina, K., Zhang, Q., Salcedo, D., et al.: Characterization of ambient aerosol in Mexico City: The organic component, in preparation, 2005 the particle phase and are thus likely exposed to similar microphysical transformations and scavenging processes, SO 2 concentration is a better indicator for SOA concentration, at least during this study Emission ratios of HOA and OOA-to-SO 2 ratio We examine here the emission ratios of HOA to primary pollutants for this study. The average ratio of HOA to EC is 1.1±0.22 µg/µgc during this study. It is similar to the average POA to EC ratio in the Northeast US estimated with a dispersion model (=1. µg/µgc) (Yu et al., 200), as well as the value estimated from emission inventory during Pittsburgh summer time (=1.2 µg/µgc; calculated from Cabada et al., 2002). The emission ratio of HOA to CO for this study, estimated after subtracting the Northern Hemisphere background of CO (0.1 ppm) from the measured CO concentrations, is.3 ng m 3 /ppbv. (The linear regression slope of HOA vs. CO with the intercept forced through zero is 3.7 ng m 3 /ppbv). These values are lower than the POA to CO emission ratio in New England (9. ng m 3 /ppbv) determined based on correlated behavior of total OA with gas-phase tracers (de Gouw et al., 2005) and the POA to CO ratio estimated from the AMS data in Tokyo, Japan (11 ng m 3 /ppbv) (Takegawa et al., 2005a ). All of these numbers are larger than the average POA to CO emission ratios measured during a tunnel study in California 1.8 ng m 3 /ppbv for diesel trucks and 0.8 ng m 3 /ppbv for light-duty vehicles (calculated from Kirchstetter et al., 1999). In addition, the emission factor of HOA to NO x of this study (i.e., 2 ng m 3 /ppbv) is roughly 3 times the POA to NO x ratios of diesel trucks and light-duty vehicles (average 16 and 11 ng m 3 /ppbv, respectively) from the California tunnel study (calculated from Kirchstetter et al., 1999). Possible reasons for the variations in these measured ratios include different sampling locations, seasons and meteorological conditions, different vehicle fleets (including different emission standards for California vehicles), as well as different measurement methods and assumptions applied for POA estimation. Since sulfate is a better indicator for OOA concentration (see discussion above), we estimate the average concentration ratio of OOA to sulfate as 0.38 (dimensionless) during this study (Fig. 3d). The significance of this number is that it provides a first order estimation of the mass concentrations of OOA based on measured sulfate concentrations in fine particles in the Pittsburgh region during Fall. A survey of the OOA to SO 2 ratios based on AMS data at many locations Takegawa, N., Miyakawa, T., Kondo, Y., Jimenez, J. L., Worsnop, D. R., and Fukuda, M.: Seasonal and diurnal variations of submicron organic aerosols in Tokyo observed using the Aerodyne Aerosol Mass Spectrometer (AMS), J. Geophys. Res., submitted, 2005a. Atmos. Chem. Phys., 5, , 2005

8 3296 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Figure Fig.. Time variations of the size distributions of (a) HOA, (b) OOA, and (c) sulfate during 7 22 September 2002 in Pittsburgh. Missing data points are due to either occasional instrumental malfunction or maintenance/calibration. in the Northern Hemisphere will be provided in a separate paper. 3.2 Size distributions of HOA and OOA Change of size distributions of HOA and OOA as function of time The image plots in Figs. a and b provide an overview of the temporal variations of the HOA and OOA size distributions during this study, showing again very different behaviors for HOA and OOA. OOA mostly resides in the accumulation mode with vacuum aerodynamic diameters (D va ; De- Carlo et al., 200) larger than 250 nm while HOA displays a much broader distribution that extends into the ultrafine mode (D va <100 nm). Typically 30% of the HOA mass is associated with ultrafine particles compared to less than 5% of the OOA mass. Note that the AMS size distributions presented here are shown vs. D va, which is the aerodynamic diameter measured under free-molecular regime flow conditions. For a spherical particle, D va equals the product of its physical diameter and density. Given that the average density of the bulk Pittsburgh particles was roughly 1.5 during this study (Zhang et al., 2005b, 200), to a first approximation, 250 nm in D va corresponds roughly to 170 nm in physical diameter for spherical particles. The size distribution of OOA tracks the behavior of sulfate (and ammonium) throughout the entire study (Figs. b and c) (also see Zhang et al., 2005b and Suppl. Info), echoing the fact that their mass concentrations are highly correlated (Fig. 2; see Sect. 3.1). Simultaneous growth of OOA and SO 2 size distributions is observed during some periods, e.g., from the afternoon of 12 to 1 September a period that follows an intense new particle formation event (see Sect. 3.5 for detailed discussion). These observations suggest that oxygenated organics are likely internally mixed with NH + and SO2 and that both OOA and SO 2 are formed over similar regional scales. The HOA size distribution pattern is distinctly different from those of OOA and SO 2. It is generally much broader, showing a pronounced ultrafine mode that increases at night and in the morning. On average, only 50% of the total HOA mass is associated with the accumulation mode. Figure 5a summarizes the average size distributions of HOA, OOA, and inorganic aerosols species of the entire study that again demonstrate the overall resemblance of OOA to secondary aerosol species (NH +, SO2, and NO 3 ). Compared to the size distributions of SO 2 and NH +, those of OOA and NO 3 are slightly broader, extending more into the Atmos. Chem. Phys., 5, , 2005

9 Figure 5. Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols 3297 smaller sizes (<300 nm). In the case of nitrate, this likely reflects active gas-particle partitioning due to its semivolatile character and the strong influences of ambient temperature and relative humidity on the partitioning (Seinfeld and Pandis, 1998). Similarly, the broader OOA distribution suggests a stronger influence of local gas-to-particle partitioning on OOA than on NH + and SO2 formation. This is consistent with the known semivolatile character of some SOA compounds (Sheehan and Bowman, 2001), compared to the nonvolatile character of sulfates. Figure 5b shows the average fractional contributions of HOA and OOA to total organic mass as a function of aerosol size, from which we estimate that 75% of the accumulation mode organic mass is OOA. In contrast, 75% of the organic mass in ultrafine aerosols is HOA Diurnal variations of the size distributions The average diurnal image plots of the HOA, OOA, and sulfate size distributions are shown in Fig. 6. As pointed out in Sect. 3.1, the dip at 17:00 18:00 EST on the diurnal plots of OOA and sulfate is primarily due to a rainfall event in the afternoon of 15 September. These figures are analogous to Fig., showing that HOA has distinctly different behavior than OOA and SO 2 and that the highest OOAto-organics fraction (and the accompanying shift of particles toward larger sizes) preferentially occurs in the afternoon. Figure 7 provides a survey of the average size distributions of HOA and OOA and their relative contributions to the total organic mass during different hours of the day in correspondence to Fig. 6. Even in early afternoon, when the HOA mass loading is the lowest and its size distribution the narrowest, the ultrafine mode organics are about 50% HOA Size distributions of HOA and OOA for periods with different HOA and OOA fractions We display in Fig. 8 the average size distributions of HOA and OOA and the correlations of HOA to CO and OOA to SO 2 during: 1) high HOA and CO periods (when both HOA and CO are in the upper 75th percentile of their absolute concentrations) that represent the situation of intense primary combustion emissions and high loading of fresh organic aerosols; 2) typical situation when the fractional contribution of OOA to the total organics (OOA%) is within the 25th 75th percentile of its absolute values (corresponding to periods when OOA contributes 58% 81% of the total organic mass); and 3) aged aerosol periods when OOA% are in the top 25th percentile of its absolute values (corresponding to periods when OOA contributes more that 81% of the total organic aerosol mass). In general, there is a clear shift of all species (including, SO 2, OOA, and most dramatically HOA) to larger mode size with higher OOA fraction (e.g., Fig. 8a-1 vs. Fig. 8c-1). The size distribution of HOA demonstrates an increasingly prominent accumulation mode Fig. 5. (a) Average size distributions of HOA, OOA, and particle phase inorganic ions (NH +, NO 3, and SO2 ) and (b) the size resolved fractional contributions of HOA and OOA to total organic aerosols in Pittsburgh over the entire study (7 22 September 2002). with higher OOA fraction. Note that although part of the narrowing of OOA distribution observed could be due to limited transmission of the AMS lens at high particle sizes (Jayne et al., 2000; and see discussions in Zhang et al., 2005b), such effect is expected to be fairly small during this study since the majority of Pittsburgh organic mass is in submicron aerosols (Cabada et al., 200b). In addition, there are several trends observed: (1) the size distributions of OOA and SO 2 are very similar under all situations; (2) HOA always dominates the composition of small particles (D va <100 nm), even during very high OOA periods (Fig. 8c-1); 3) the linear regression slope of HOA vs. CO is somewhat lower with higher OOA fraction (Figs. 8b-1 8b- 3); and ) the correlation of OOA to SO 2 is always good but the OOA/SO 2 ratio decreases slightly with high OOA fraction (Figs. 8c-1 8c-3). In addition, while not shown here, the correlation of OOA with O 3 does not improve (i.e., r 2 0) at high OOA. 3.3 Mass spectra and estimated elemental compositions of HOA and OOA Together with the mass concentrations of HOA and OOA, complete mass spectra of these two components were extracted using the deconvolution technique described in Zhang et al. (2005a). Because of the clear separation of the HOA signals from the OOA in measured mass spectra, we are able to estimate the possible elemental compositions of each m/z in the HOA and OOA spectra and thus the elemental composition of the organic aerosol. Table 2 lists the estimated compositions of the 1 and 16 most abundant m/z s, Atmos. Chem. Phys., 5, , 2005

10 3298 Figure 6 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Fig. 6. Average diurnal variations of the size distributions of (a) HOA, (b) OOA, and (c) sulfate during 7 22 September 2002 in Pittsburgh. The raw data have been averaged into 20 min intervals. accounting for 75% and 67% of the OOA and HOA signals, respectively. The assumed compositions of the major peaks in the spectra were verified by examining preliminary data on the organic mass spectra of ambient aerosols acquired by a high-resolution ToF-AMS 5. For m/z s not listed in the table, we assume that those of HOA have the same average C:H ratio as the average of 16 HOA m/z s listed in Table 2 (i.e., average molecular composition is (CH 2 ) n ) and those of OOA have an average C:H:O ratio same as the average of the major OOA m/z s in Table 2 excluding 17, 18, 28 and (i.e., average molecular composition is (C 2 H 3 O) n ). These elemental compositions are first order estimations since we only included C, H, and O atoms. The omission of nitrogen atom may influence the OM:OC estimates since nitrogen-containing organic compounds have been detected in ambient aerosols (Li and Yu, 200; Zhang and Anastasio, 2003; Zhang et al., 2002a, b]. However, the influence is expected to be relatively small because C, H, and O are 5 DeCarlo, P., Aiken, A., Jimenez, J. L., et al.: A High- Resolution Aerosol Mass Spectrometer, in preparation, the three dominant atoms reported in aerosol organic species (Seinfeld and Pandis, 1998). Recent studies reported that N atoms typically account for 10% or less of the total organic mass in atmospheric fine particles and fog waters (e.g., Zhang and Anastasio, 2001; Zhang et al., 2002a). In addition, by using mass spectra to derive the elemental composition of molecules (Table 2) we assume that the elemental composition of the ions is on average the same as the elemental composition of the parent molecules. This assumption could introduce some bias on the estimated elemental composition if certain functional groups or molecular structures have a greater tendency to end up as either ions or neutrals in the fragmentation process. Figure 9 shows mass spectra of HOA and OOA colored with the contribution of C, H, and O at each m/z. See Zhang et al. (2005a) for detailed discussion on these two mass spectra; only the major points are summarized here: 1) the HOA spectrum demonstrates prominent ion series characteristic of hydrocarbons and shows remarkable similarity 6 to the measured AMS mass spectra of diesel exhaust aerosols and lab-generated lubricating oil and diesel fuel aerosols Atmos. Chem. Phys., 5, , 2005

11 Q. Zhang Figure et al.: 7. Hydrocarbon-like and oxygenated organic aerosols 3299 Fig. 7. Average size distributions of HOA, OOA and sulfate and the size resolved fractional distributions of HOA and OOA to total organics during different hours of day. Table 2. Estimated elemental compositions of the major m/z s (total number=270) in HOA and OOA a. HOA OOA m/z % Sig. b m/z Comp. m/z % Sig. b m/z Comp C 2 H CH C 2 H HO C 3 H H 2 O 2 1. C 3 H C 2 H C 3 H CO C H CHO C H CH 3 O C H C 2 HO C 5 H C 2 H 2 O 69.9 C 5 H C 2 H 3 O C 5 H CO C 6 H CO 2 H C 6 H C 3 HO C 6 H C 3 H 3 O C 7 H C 7 H 13 rest of m/z s 33 (CH 2 ) c n rest of m/z s 25 (C 2 H 3 O) d n a The mass spectra of HOA and OOA (up to m/z=150) are presented in Fig. 7 and those in logarithmic scale (to show low signal m/z s more clearly) are presented in Fig. 11 in Zhang et al. (2005a). b % of the total signals in each component (HOA or OOA) mass spectrum that was detected at the specified m/z. c Average molecular composition assumed for the rest of the HOA m/z s. d 7 Average molecular composition assumed for the rest of the OOA m/z s. Atmos. Chem. Phys., 5, , 2005

12 3300 Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Figure 8 Fig. 8. Average size distributions and the size resolved fractional contributions of HOA and OOA to total organic aerosols during: (a-1) High HOA (above 75th percentile of HOA concentration) and CO (above 75th percentile CO concentration) periods; (b-1) Periods when the OOA to total organics ratios (OOA%) are within the 25th 75th percentile of its value and (c-1) Periods when OOA% are in top 75th percentile the value. To the right of the size distribution plots are the scatter plots and linear regressions between HOA and CO (a-2, b-2 and c-2) and OOA vs. sulfate (a-3, b-3 and c-3) during the corresponding periods. Red lines are the linear fits to the data. All the linear fits were performed with intercept forced through the origin. 8 Atmos. Chem. Phys., 5, , 2005

13 Q. Zhang et al.: Hydrocarbon-like Figure 9. and oxygenated organic aerosols 3301 Fig. 9. Mass spectra of (a) HOA and (b) OOA, colored with the estimated contribution of each element (C, H, and O) to the mass of each m/z fragment. The elemental compositions of each m/z in HOA and OOA are estimated according to Table 2. Figure 10. (Canagaratna et al., 200); 2) the OOA spectrum is dominated by m/z (CO + 2 ) and m/z 28 (CO+ ) and demonstrates close similarity in the overall pattern with those of aged/oxidized organic aerosols in rural and urban areas; and 3) the OOA spectrum is also qualitatively similar to the AMS mass spectrum of Suwannee River fulvic acid (Alfarra, 200), which is a class of highly oxygenated organic compounds that have been proposed as models of the highly oxidized organic aerosols that are ubiquitous in the atmosphere (Decesari et al., 2002). In addition, neither HOA nor OOA mass spectrum represents individual species, but rather, they represent mixtures of many individual organic species associated with the same group of sources and atmospheric processes (i.e., urban emissions vs. regional secondary aerosol). Based on estimated elemental compositions of m/z s, we estimate that the average molar ratio of C:H:O in OOA is 1:1.6:0.8 (or 5:8:) and that the average molar ratio of C:H in HOA is 1:1.9 (or 10:19). The organic mass to organic carbon ratios (OM:OC) of HOA and OOA are estimated at 1.2 and 2.2 µg/µgc, respectively. This HOA OM:OC ratio is consistent with the value (1.2 µg/µgc) of hydrocarbons (Turpin and Lim, 2001) the major components of urban fresh combustion aerosols. In addition, the OOA OM:OC ratio is close to the value estimated for nonurban aerosols (2.1±0.2 µg/µgc) (Turpin and Lim, 2001) but is significantly higher than estimates based on functional group measurements by FTIR spectroscopy for samples collected in northeastern Asia and the Carribean ( , mean 1. µg/µgc) (Russell, 2003). The average OM:OC ratio of submicron organic aerosols (OOA plus HOA) estimated with this procedure is 1.8, a 9 Fig. 10. Scatter plot between organic carbon concentrations estimated from AMS mass spectra and component-specific m/zelemental compositions, and those measured by the Sunset Lab carbon analyzer. value that is close to the number determined by comparing organic mass concentration from the AMS and organic carbon concentration from a Sunset labs carbon analyzer (Zhang et al., 2005b). It is also comparable to the number (1.6±0.2) proposed by Turpin and Lim (2001) for urban aerosols. This analysis is summarized in Fig. 10, where the organic carbon contents derived from the HOA and OOA mass spectral analysis show good agreement with the organic carbon (OC) concentrations from the carbon analyzer (r 2 =0.87 and the linear regression slope=1.01±0.11). Atmos. Chem. Phys., 5, , 2005

14 Figure Q. Zhang et al.: Hydrocarbon-like and oxygenated organic aerosols Fig. 11. (a) and (a ) Time series, scatter plot and linear regression between (a) & (a ) hydrocarbon-like organic carbon (HOC) concentrations from the AMS measurements and primary organic carbon (POC) concentrations estimated from the EC measurements assuming a POC to EC ratio of 1 and (b) & (b ) oxygenated organic carbon (OOC) concentrations and secondary organic carbon concentrations (SOC=OC POC). AMS HOA and OOA data were reduced to 2 h averages according to EC/OC measurement time intervals. Missing data during 11 September 15 September were due to malfunction of the Sunset Laboratory carbon analyzer. The linear regression parameters and r 2 s are shown in the scatter plots. 3. Comparison with results from previous studies 3..1 Comparison with estimates from the EC/OC tracer method The elemental carbon (EC)/organic carbon (OC) tracer method has been frequently used to estimate the carbon concentrations of primary and secondary organic aerosol (POC and SOC, respectively) (Cabada et al., 2002, 200a; Castro et al., 1999; Park et al., 2005; Polidori et al., ; Turpin and Huntzicker, 1991, 1995). This method derives the POC concentration based on the EC measurements assuming a constant POC to EC ratio (Turpin and Huntzicker, 1991, 1995). SOC is subsequently estimated as the difference between measured total OC and the estimated POC based on the assumption that SOA is formed through gas to particle conversion that involves no EC emissions. Note that the POC and SOC concentrations thus estimated may contain significant uncertainties due to 1) the operational definition for the OC and EC fractions in thermal-optical analysis (Gelencser, 200; Turpin et al., 2000); 2) the uncertainties associated with the estimated POC/EC ratios for the average of the combustion emission sources (Turpin and Lim, 2001); Atmos. Chem. Phys., 5, , 2005 and 3) variations in time of POC/EC ratios due to factors such as varying fractions of diesel and gasoline vehicles on the road (Harley et al., 2005). Despite these limitations, the EC/OC method has been applied frequently because of the lack of direct measurement techniques that can distinguish POA from SOA (Kanakidou et al., 2005). The POC and SOC concentrations during this study are estimated using the PM 2.5 EC and OC data from thermaloptical transmittance carbon analysis (Polidori et al., ): POC=1 EC and SOC=OC POC. The POC/EC ratio of 1 was estimated based on summertime emission inventories in Pittsburgh (Cabada et al., 2002). Figure 11 compares the POC and SOC estimates to the concentrations of hydrocarbon-like and oxygenated organic carbon (HOC and OOC, respectively) estimated according to their estimated molecular compositions (see Sect. 3.3). Overall, HOC correlates well to POC (r 2 =0.69, Figs. 11a and 11a ) and OOC correlates with SOC (r 2 =0.52; Figs. 11b and 11b ). These correlations are consistent with the diurnal variation patterns, mass spectra, and size distributions of HOA and OOA, which all corroborate the hypothesis that most or all HOA is POA and that most or all OOA is SOA. 51

15 Q. ZhangFigure et al.: Hydrocarbon-like 12 and oxygenated organic aerosols 3303 (a) AMS estimates (b) POC:EC = 1.2:1 (c) POC:EC = 1:1 (d) POC:EC = 2:1 OOA (66%) HOA (3%) SOA (6%) POA (36%) SOA (71%) POA (29%) SOA (7%) POA (53%) Fig. 12. Fractional distributions of (a) HOA and OOA estimated from the AMS data and POA and SOA estimated from OC/EC measurements assuming POC-to-EC ratio (b)=1.2, (c)=1 and (d)=2. POA and SOA are converted from POC and SOC assuming OM:OC ratios of 1.2 µg m 3 /µgc m 3 and 2.2 µg m 3 /µgc m 3, respectively. The linear regression fit to HOC vs. POC has a slope of 1.33±0.27 µgc/µgc) and an intercept of 0.10±0.36 µgc m 3 (Fig. 11a). In contrast, the fit to OOC vs. SOC yields slope=0.6±0.17 µgc/µgc with intercept of 0.52±0.29 µgc m 3. The slopes of HOC vs. POC and OOC vs. SOC obtained with the intercept fixed at zero are 1.23±0.22 and 0.7±0.35, respectively. Note that perfect agreement between these two estimates of POC and SOC is not expected because they were estimated by completely different methods under different assumptions. In addition, uncertainties associated with both measurements and data analysis procedures may also contribute to the observed discrepancy. For instance, there appear to be some changes in the correlation patterns after 11 to 1 September a gap of missing POC and SOC data due to a major component failure of the EC/OC analyzer (J. Cabada, Tecnológico de Monterrey, personal communication). As pointed out at the beginning of this section, the soundness of the OC/EC method for predicting POC and SOC is strongly influenced by the choice of the POC to EC ratio and the validity of the assumption that this ratio is relatively constant during the time period of interest. To illustrate this first point, we compare in Fig. 12 the fractional distribution of POA and SOA obtained from the AMS data to those obtained from the EC/OC tracer method using different POC to EC ratios: 1) POC/EC=1.2, which is the average HOC to EC ratio estimated from this study; 2) POC/EC=1, which is estimated based on emission inventory for Pittsburgh in the summer (Cabada et al., 2002); and 3) POC/EC=2, which is approximately the average of a range of ratios estimated based on measured OC/EC ratios during periods dominated by primary emissions and with low O 3 (Cabada et al., 200a; Polidori et al., ). POA and SOA are estimated from the POC and SOC results of the EC/OC method assuming OM:OC ratios of 1.2 µg m 3 /µgc m 3 and 2.2 µg m 3 /µgc m 3, respectively (see Sect. 3.3). As shown in Figs. 12a and b, the fractional distribution of POA/SOA estimated from EC/OC measurements agrees very well to that determined from the AMS data when POC/EC=1.2 is used. In comparison, assuming POC/EC=1 yields higher fraction of SOC (Fig. 12c) while assuming POC/EC=2 projected from ambient measurements leads to significantly less SOC and twice more POC (Fig. 12d). Note that POC/EC=2 may be a significant overestimation since fine particles in Pittsburgh are strongly influenced by regional sources and thus contain a relatively high background of oxidized organic species (Anderson et al., 200; Tang et al., 200; Zhang et al., 2005b). Even during periods with intense primary emissions and reduced mixing (e.g., morning rush hour) OOA contributes more than 50% of the total organic mass on average (Fig. 1). In addition, a low concentration of O 3 does not necessarily imply that SOA is also low because O 3 is a much shorter-lived photochemical product than SOA. O 3 can be titrated away quickly by NO emitted by traffic, while SOA will persist. Figure 3d, for example, shows that the r 2 between OOA and O 3 is almost zero during this study (see Sect ). In fact, based on the OOA/HOA ratios observed during the morning rush hour ( 1.3:1), we estimate that the POC/EC assumptions projected from ambient measurements in Pittsburgh may be biased high by up to a factor of 2, suggesting that POC/EC 1 is a better estimate. In addition, these comparisons, together with the good correlation between HOA and EC, also indicate that the EC/OC tracer method can produce a useful estimation of POC and SOC as long as the correct POC/EC ratio is given. Otherwise, the POC and SOC estimates can have considerable errors if this ratio is not well constrained (as is often the case in practice) Comparison with results from VOC-based factor analysis Millet et al. (2005) also estimated the fraction of secondary carbon in Pittsburgh aerosols during July August 2002, based on a new hybrid source apportionment method incorporating elements of the EC/OC tracer method and of joint factor analysis of a large set of VOCs. These authors estimate an OC/EC ratio of 1.36 for primary combustion 52 emissions, in qualitative agreement with our results discussed in the previous section. They also estimate that secondary organic carbon contributed 37% of the total carbon, while 35% of the carbon was classified as regional background. Atmos. Chem. Phys., 5, , 2005